Datasets:
Add data integrity statement about NaN for missing hand tracking
Browse files
README.md
CHANGED
|
@@ -23,6 +23,8 @@ pretty_name: Dynamic Intelligence - Egocentric Human Motion Annotation Dataset
|
|
| 23 |
|
| 24 |
RGB-D hand manipulation dataset captured with iPhone 13 TrueDepth sensor for humanoid robot training. Includes 6-DoF hand pose trajectories, synchronized video, and semantic motion annotations.
|
| 25 |
|
|
|
|
|
|
|
| 26 |
---
|
| 27 |
|
| 28 |
## 📊 Dataset Overview
|
|
|
|
| 23 |
|
| 24 |
RGB-D hand manipulation dataset captured with iPhone 13 TrueDepth sensor for humanoid robot training. Includes 6-DoF hand pose trajectories, synchronized video, and semantic motion annotations.
|
| 25 |
|
| 26 |
+
**Data Integrity:** We preserve raw observability—when hand tracking fails (out-of-frame, occlusion), states are explicitly marked as `NaN` rather than interpolated. This enables downstream models to reason about uncertainty instead of learning from fabricated data.
|
| 27 |
+
|
| 28 |
---
|
| 29 |
|
| 30 |
## 📊 Dataset Overview
|